WHAT CHANGED

Changelog

What changed in the index and the site, newest first. Entries mentioned in a release link to their detail pages.

Jevland v1.4 2026-10-05

From discovery to first run

Expanded the production catalog from 542 to 581 entries: 37 projects with repositories that had at least 500 stars at selection, including ten with at least 1,000, plus Needle and Ripple. Referenced GitHub addresses rose from 366 to 404; shared repositories and aliases are counted separately from projects, and stars belong to the whole repository.
Published four practical guides: connect with Python, select relevant context, connect with JavaScript/TypeScript, and route an agent’s next step. Each includes dated references, downloadable examples, Copy code, setup instructions and troubleshooting.
Connected existing task collections to ten project-specific How to run sections and related guides, with stack, requirements, credentials and dated setup references.
Added Save and Unsave on project cards and detail pages, a Saved list and text-link export. Lists stay in this browser at the current site address; production and preview have independent saved lists.
Separated HTTP availability, browser demo scenarios, build and live API observations. Each recorded check shows its scope and date; absent observations show Not checked.
Improved mobile repository columns and navigation, keyboard access and text contrast. Deduplicated repository star totals, added canonical URLs and share cards, reduced homepage data, and repaired the Akita article destination.

Records touched: Sub2API: native System One gateway, OpenHuman: Jev tool-search ranker, Oh My ClaudeCode: opt-in Jev judgments, AgentScope: Jev classifier model, Composio: TypeSafe tool decisions, GPT Researcher: Jev context selection, DeepEval: structured Jev evaluation, Opik: TypeSafe evaluation tracing, json-render: experimental Jev composition, OpenCodex: Jev model routing, TradingAgents: Jev social-post screening, AI Hedge Fund: native Jev adapter, QuantDinger: Jev pre-trade decisions, MemSearch: Jev memory reranking, AgentConnect: typed agent decisions, LLPhant: PHP Jev classifier, ReqLLM: Elixir TypeSafe evaluation, Agent!: optional Jev action advisor, Surf CLI: bounded Jev browser commands, Photoshop MCP: Jev intent routing, ThoughtDAG: Jev conversation recall, Valen: independent multimodal decision model, JevHarness: authored decision pipelines, Hippo Memory: local memory with Jev ranking, OpenMausBot: Jev agent selection, OpenMuse: Jev decision controls, Pi extensions: TypeSafe decisions and recall, Needle: semantic search inside webpages, Ripple: linked edits in Google Docs, GenOffice: Jev file-search ranking, Agent Squad: typed Jev routing, Pro Workflow: advisory System One checks, Math-To-Manim: Jev checkpoint review, XERJ: Jev judgments over search results, PyGPT: inline System One decisions, Paca: Jev project and task routing, Pi MCP Adapter: Jev semantic tool search, GoModel: native Jev gateway provider, OpenHarness: Jev Sheets question trials.

Build paths & saved projects 2026-10-05

From a task to a project and a first run

Added JavaScript/TypeScript and agent next-step guides with downloadable examples and offline SDK checks.
Added dated How to run references for ten projects and linked them from existing task collections.
Separated HTTP, browser scenario, build and live API observations on project pages. Added browser-local saved projects and text export.

Guides & site usability 2026-10-04

Copyable walkthroughs and clearer browsing

Improved both guides with separate Windows and macOS/Linux commands, Copy code, Python highlighting, reading-time estimates, troubleshooting and links from related projects and collections.
Fixed mobile repository columns, made result counts accessible and removed duplicate stars from repository totals. Added compact mobile navigation, canonical URLs and share images.
Reduced the homepage's embedded availability data while preserving its checks and filters, and restored the full Akita article link.

Guides preview 2026-10-04

Two practical guides for a first integration and context selection

Added a Guides section with original walkthroughs for connecting Jev through the Python SDK and selecting relevant passages before generation.
Both guides include official references, dated reviews, catalog examples and downloadable Python scripts. SDK request and response handling was checked offline; live inference and example thresholds are not presented as measured results.

Batch 18 2026-10-04

Thirty-seven repositories and two document tools

Added 37 new repositories with at least 500 GitHub stars at the review snapshot, including ten additional integrations with at least 1,000 stars (581 entries total).
Added Needle for semantic webpage search and Ripple for related edits in Google Docs. Their shared parent repository is indexed once; the selected demo preview gallery remains 15.
Expanded the preview with office file search, multi-agent routing, coding advice, animation review, search judgments, desktop decision tools, project automation, MCP tool selection, an API gateway and frozen-row spreadsheet trials.

Batch 17 2026-10-03

Fifty reviewed repositories and dated availability checks

Added 50 repository projects covering code search, agent routing, semantic lint, released decision weights, serving, independent evaluation and applied interfaces (542 entries total).
Added an Interfaces & creative workflows collection and refreshed the current briefing. New descriptions distinguish live inference, local fallback and recorded simulation; the selected screenshot gallery remains 15.
Added dated availability statuses, check filters and contextual problem reporting through Tally. Daily HTTP checks cover 502 unique destinations; selected PlayJev and minojev saved-data scenarios run separately in Chromium.

Records touched: Oko: local code search and prompt-time retrieval, jevr: BM25 recall and Jev reranking for code and documents, Evoke: typed plans for installed reflex programs, LangChain Skill Router: per-turn skill selection with Jev, oxlint-plugin-jev: plain-English rules over source code, patdown: Markdown rule linting and agent write checks, pi-heed: conversational constraints for Pi tool calls, Fastbrowse: browser tasks with typed decisions and citations, xpressabhi Jev Browser: a harness-independent decision core, forvela Jev Agent Browser: bounded browser execution for parent agents, Jev Agent Orchestrator: Databricks catalog-based agent routing, Jeview: a local proxy and inspector for Jev decisions, every: ask a yes/no question of each source function, xafold Jev Router: cache-aware model routing for Claude requests, Jev Model Router: turn-level model and effort hooks, Jev Agent Skill Router: bounded selection with abstention, Jev Agent Tools: bounded conversation context for Pi and Codex, Jev Agent Desk: local control of Claude and Codex sessions, JevK5: open decision weights with CUDA and GGUF runtimes, JevAny: train and deploy decision models across backbones, Eikos: open typed decisions for financial rules, EdgeJev: export and quantize decision models for CPU, Openlayer jevals: batched decision-model evaluations and guardrails, wellposed: lint the structure and meaning of Jev questions, AnyDecisionModel: typed decisions from Swift sessions, sys1: a Rust server for open System One models, JEVfire: parallel finite-choice decisions with vLLM, ASSAY-001: calibration and type-safety evidence on two intent corpora, Jev × LexGLUE: typed decisions across seven legal tasks, Jev reranking: saved outputs across retrieval tasks and scoring strategies, jev-fanout-bench: measure batching, billing and answer stability, System One Playground: compare decision engines on Apple Silicon, Jev Does Not Play Dice: inspect probabilities on known uncertainty, AgentJev: a small candidate-head model for agent decisions, Shapeshift: text input that becomes a useful interface, EmbodiedJev: an inspectable robot decision workbench, Jev Voice: local speech and bounded Mac actions, Refgarden: spatial reference search with bounded choices, RoboJEV: two-stage control in MuJoCo, Reflex: a simulated drone autonomy lab, Jev Grand Prix: racing lines and pedals, DoomSat: a game payload through a mission-control stack, IntelliPrompter: check off talking points as you speak, Discoprint: a classified discography in the terminal, Semantic Live Caption: annotate a speech transcript, Askable Arm: English goals over a robot skill catalog, Read with Jev: an emotion-aware novel reader, Roverlab: an inspectable planetary expedition sandbox, PROMPT FPS: browser-native Doom control, SmartSwitch: predict the next Mac app.

Batch 16 2026-10-02

Thirty checked repositories, seven more demo previews and three research papers

Added 30 repositories covering browser and agent tools, routing, code search, independent decision models, evaluation, games and simulations, plus three primary research papers (492 entries total).
Added seven attributed demo screenshots, bringing the selected gallery to 15. Recorded runs, local interfaces and hosted demos keep their original evidence context; new repositories join the existing GitHub monitor.

Records touched: tontoko Jev Browser: Playwright operations and semantic assertions, legostin Jev MCP: Chrome control and decision traces, Computer Use Jev: macOS Accessibility automation, hyspacex Jev Router: session model admission, claude-jev: rule checks and verbatim compaction, pi-jev-auto-mode: typed tool permission gate, genigrep: source evidence packs for coding agents, Jev Navigator: code relationships and semantic search, Jev Playwright: change-aware browser test selection, Jev × WebMCP: Chrome tool schema routing, Gatekeeper: YAML rules for agent and skill selection, burnigtm Jev MCP: exact prepared call routing, Luce: train and calibrate a decision model for your task, JevForge: decision-data synthesis, training and web-action scoring, Open-Jev: trained decision heads and an inspectable replay workbench, lev: a Qwen 4B decision model with dual readouts, OpenDecision: local typed decisions with document evidence, poorjev: local typed decisions with calibration and abstention, jevos: TypeSafe-compatible decisions on a laptop CPU, JevBench: an independent benchmark for typed decision models, jevcal: fit confidence thresholds and detect decision drift, jev-acento: a paired Spanish-language accuracy and calibration audit, JevBall: 22 football players with inspectable decisions, HEIST//ONE: typed guard judgments in a browser stealth game, TerraBlind: Jev boss-fight intent over Terraria controls, Jev Autopilot: typed drone stick controls in a generated city, JevsBistro: replayable restaurant service decisions, Soupbase: bilingual lateral-thinking puzzles judged by Jev, Jev Playground: typed plans rendered as sheet music and MIDI, Little Airways: aircraft and tower judgments in a toy airspace, Evaluating Jev across 37 public datasets, Jev in the Wild: a study of the emerging project ecosystem, NumericJev: numerical decoding through structured range choices.

Scheduled GitHub checks 2026-10-02

Automatic repository metadata checks

Referenced GitHub repositories are checked on a six-hour schedule. Stars, language, repository status and last-push dates refresh from successful observations.
Private checkpoints preserve progress across retries and API limits. Failed requests retain the last successful metadata, and Updates and Repos show individual check dates.

GitHub metadata freshness 2026-10-01

GitHub metadata freshness and scheduled-check preparation

Updates and Repos now distinguish successful metadata checks from later failed attempts and show the age of saved observations. Repository sorting uses full push timestamps.
Prepared a resumable collector, conditional GitHub checks and a private-state preview workflow planned every six hours. Scheduled checks remain pending until configuration is complete.

Updates and demos 2026-10-01

Repository activity and attributed demo previews

Added the latest recorded GitHub push for each repository, individual metadata check dates, search and sorting. Redirected aliases are combined; missing metadata is excluded from activity.
Eight selected demo previews now have a gallery, small local images, source attribution and direct demo links. Recorded runs keep their context and evidence labels.

Batch 15 2026-10-01

Twenty-nine checked repositories and projects

Added 29 public repositories covering typed-decision clients, code search, agent and browser tools, games, evaluation and independent local runtimes (459 entries total).
Every addition has a checked repository destination and original source-backed notes; equivalent names and existing destinations were screened before import.

Records touched: System One for TypeScript: provider-neutral decisions, Stumble jev-go: Go SDK and interactive CLI, go-jev: Go client and UNIX decision pipelines, jev-rs: Rust System One scoring and evaluation engine, jevcompat: System One API conformance suite, larguesa jev-search: semantic line retrieval, JevGrep: semantic source retrieval through CLI and MCP, JevFind: two-stage semantic code finder, pi-jev-tools: ranking and retrieval tools for Pi, jevgrep jg: Rust natural-language code search, jev-browse: Jev-driven Chrome automation, Jev Browser: Playwright agent through MCP, CLI or library, dsh-jev: Jev decision layer for DeepSeek Harness, clash-jev: live Clash Royale decisions and recorded replays, PlayJev: a 0.8B visual decision model for ten games, Jev × LIBERO: robot decisions with local physics previews, Jev plays StarCraft: a recorded Strongarm mission, Jev × Civilization II: an experimental campaign harness, Component Charades: Jev guesses a UI component, Coffee Under Fire: Jev-controlled NPCs in a browser shooter, Jev Lab: a 100-NPC town and a shogi opponent, WindowsJev: bounded Windows actions and durable research, JevTest: exploratory browser tests with replayable evidence, Footwork: Jev guards and mechanical computer-use steps, system-one: calibrated local models on Apple MLX, Ollaya: pull and serve open decision models locally, minojev: a trained decision head on a frozen language model, YC × Jev: startup search and a reranking benchmark, Jevals: labelled benchmarks and per-decision logs.

Site update — 2026-09-29 2026-09-29

Easier discovery, a briefing and a suggestion channel

Added direct routes into six use-case collections, a three-record Start here path and a dated briefing of recent additions.
Opened a suggestion and correction form; submissions are reviewed for sources and duplicates before any catalog change.

Batch 14 2026-09-29

Seventeen checked projects, demos and Jev references

Added six previously unlisted GitHub repositories, a Cua model release, a live people-search workflow, a guide directory, two articles, an X-only search demo and five source-attributed Jev roundups (430 entries total).
Existing repositories and unrelated general-tool links from the supplied X posts were not duplicated; project destinations and X evidence remain separate.

Records touched: Twelve agentic uses for Jev, Twenty proposed Jev and open-source tool combinations, Beacon: cross-harness memory for coding agents, jevgrep: semantic code search for coding agents, Made with Jev: guide index, AnyJev: typed decisions from open models, Jev in games, browsers, trading and context tools, JevPilot: browser driving simulator, Voice Browser: spoken commands to bounded browser actions, Jev Trade: Hyperliquid trading bot, A Jev market-making architecture, Ten Jev repositories by use case, Cua-S1-4B-0.2: open computer-use decision adapters, Y Combinator startup search with Jev, Jev and Codex: a Japanese implementation guide, Treg people search with a Jev screening step, Ten Jev repositories and their roles.

Batch 13 2026-09-28

Eight source-backed repositories and projects

Added eight checked public repositories: clients for TypeScript, Java and Go, a Chrome extension, a local workbench, demos, an agent skill and a research harness (413 entries total).
Each repository is both the evidence source and primary destination; no unverified X post was attached.

Records touched: system-one: Promise and Effect clients for typed decisions, Spring AI TypeSafe: Java SDK and Jev integrations, kataras/jev: a Go client for System One, Jev for Chrome: in-tab browser actions, Jev Workbench: versioned judgment functions, sawzhang/jev-demo: five demos and Jev measurements, Jevify: an agent skill for typed-decision workflows, Community jev-harness: proposal review with four questions.

Batch 12 2026-09-26

10 projects and repositories from verified X sources

Added ten source-backed entries from checked X posts and public project or repository pages (405 entries total).
The batch covers meetups, code search, desktop chat, ranking, video editing, agent tools and an independent decision model; X sources remain separate from project destinations.

Records touched: Shibuya Machimachi: a Jev-assisted meetup app, CodeNib: Jev reranking for code search, Jev Chat for Windows: ranked WeChat reply drafts, HN Jev Scores: adjustable Hacker News ranking, jevclip: transcript-driven video highlights, mizchi/jev-playground: MoonBit client and experiments, DSH Jev Effort Selector, Agent Chaperone: MCP tool screening, CLM: an independent decision model, Caliber ai-setup: optional Jev transcript compaction.

Batch 11 2026-09-23

18 projects and repositories from verified X sources

Added 18 source-backed entries from exact X posts and publicly reachable project or repository pages (395 entries total).
The batch includes agent and browser tools, data integrations, games, a directory and an independent local decision model; original X posts remain separate from project destinations.

Records touched: qc-use: Jev-assisted UI tests from plain-language steps, rerank-bench-jev: Jev versus Qwen3 Reranker, jev-ui-map: typed actions for browser tasks, Jev decisions in BigQuery through Cloud Run, ZeroNSFWBot: optional Jev text moderation for Telegram, jevdemo: Choice, Score and Noul in Python, gmail-jev: label Gmail threads with Jev, jev-linkedin-slop-filter: label LinkedIn posts, jev-chat-jarvis: Android reply assistant, jev-chat-jarvis-mac: WeChat reply assistant, JEV DataOps: screen and evaluate training data, Jev Voice: typed decisions for Mac voice control, xo-jev: Jev plays tic-tac-toe, ASO Workbench: Jev-assisted keyword analysis, jev-voice-games: voice-driven Godot mini-games, cmd-mod-jev-nudge: decide when an agent should continue, Jev Use Case Tracker, Qwen3.5 0.8B decision model: a local Jev alternative.

Batch 10 2026-09-22

29 projects and repositories from the owner's list and X

Added 29 projects and repositories: six new finds from the owner's list and 23 from posts published on 21–22 September (377 entries total).
Updated Kev to cover its 0.6B, 4B and 8B family and linked the Hacker News discussion.

Records touched: pi-typesafe-jev: Jev as five native Pi tools, opencode-jev-orchestrator: cheap sticky model, Jev escalates hard turns, jev-guard: auto mode for every coding agent, litjev: /v1/systemone on top of a local Qwen via vLLM, Treg + Jev: three workflows, $8K/month saved, duckdb-jev (prasanthj): native, batched, cached, doom-ai-overlord: Doom on a 421M local decision model, jev-lab: 59 open-source projects around one question, Jev Desktop: open-source computer use, ~10x faster than native, openjev.cpp: a llama.cpp fork for Jev-like inference, jev_robot: a moving-target pick-and-place in MuJoCo.

MVP v1.3 2026-09-22

Cloudflare launch and reach sorting

Production deployed to Cloudflare Pages at jevland.mnimiy.com with the existing design and logo unchanged.
Added a shareable By reach sort backed by structured xViews counters; entries without a readable counter follow measured entries.
Captured a dated reach snapshot for 218 of 220 records with an X source and exposed the count on cards and detail pages.
Added a zero-dependency refresh-x-reach command, schema validation and regression tests for reach data.

Batch 9 2026-09-21

35 entries from 24 hand-picked posts and the repositories they link

35 records added (348 total). Repositories: DocJev (LlamaIndex founder; 6x faster than gpt-5.6-luna at equal accuracy), Jev-cu (Codex computer use, 580k views), third-hand (computer use with no LLM), hermes-jev-skills, jev-sift, winnow, SemDecide, itsmostafa/typesafe-mcp, Agent Desktop, OneVOneJev, prism-liquidity-agent, jev-curate, Canny, jev-browser (Ying-Kai Liao), Reticle, open-alternative-jev, y0usaf/pi-jev, AnotiaWang/awesome-jev.
Demos and products: classifier.dev, JevForm, predictive spreadsheets, video clipping, the $250-to-$25 SEO/GEO run, Jevmaxxing for marketers, CAPTCHA arbitrage (medium risk).
Writing: Jev 101, Jev use cases explained, the Jev + Claude Code stack, Jev in front of Grok Build, LLMs vs Jev explained, two Chinese project lists, the RL-verifier take on Jev-as-a-judge, DataChaz's ten-step digest, the Chinese four-step setup post.
SimpleJev.ai, killmyidea and the official skills repo gained their X posts, project site and repository respectively. Snapshot date moved to 2026-09-21.

Records touched: DocJev: document classification and splitting with Jev, Jev-cu: the fastest computer use inside Codex, third-hand: computer use with no LLM at all, hermes-jev-skills: routing, memory, compaction, GUI clicks for Hermes, classifier.dev: a free open classifier pitched against Jev, Reticle: machine-native runtime perception for web and desktop apps.

Batch 8 2026-09-20

67 projects and repositories from a wide X sweep

67 records added (313 total) from nine X searches (Latest and Top, engagement-filtered), every repository opened and confirmed. Open models and replicas: Bespoke Nimble, GitHub Next's LocalJev, Laya (plus MLX and Core ML ports), openJev-verdict 2.0, open-jev typed engine, razorback16/openjev, djev-spark, OpenThai-SystemOne, granite-decisions, jev-visual, Jeff-1, a Japanese ModernBERT, Mapika/decider-2b, a llama.cpp jev branch, the vLLM PR and the open-jev-laya-bench dataset.
Agents and coding: jev-use (Mac harness on the accessibility tree), Stanley, jev-preflight, pi-jev-guard, abide, pi-jev, a Codex model-and-effort router, Hermes plugin, an MCP server, quackd for robots, a Minecraft agent, two chess projects.
Tools and integrations: jev-align (GEPA calibration CLI), Sift, Tab Bouncer, Page Verdict, jev-lexer, Sniff Test, Jev Detector, findme, hunk-triage, jev-reranker, ielab llm-rankers calibration check, semgate (Go middleware + CTF site), Zammad and Redmine plugins, triage-bot, jevlogs, transcript-lens, hafiza-os, three trading guards (high-risk), Swift ×2, Scala ZIO and AX clients, an 'imposter Jev' gateway benchmark.
Writing and directories: JEV HUB (216 videos / 214 articles), awesome-jev-use-cases, jev-llm.com, 'You could have built Jev', Yonyon's three-days-in-production notes, Akita's counter-take, LangChain's harness webinar, OpenRouter's playground note, Greptile realtime.
Jev as a Judge now points at LangChain's own announcement (230k views) as the primary source.

Records touched: Bespoke Nimble: open data, open model, open recipe, LocalJev by GitHub Next: a Jev-compatible local API, Laya: the open decision model that predates Jev, jev-align: calibrate Jev to your decision criteria with GEPA, jev-use: a Mac computer-use harness without a vision model, Stanley: a Jev-first, self-evolving coding agent, semgate: semantic request filtering for Go HTTP, JEV HUB: 216 demo videos and 214 long posts, categorised.

Batch 7 2026-09-20

16 entries from the owner's X bookmarks; GitHub metadata complete; new badge retired

16 records added (246 total) from bookmarked X posts: TipTour (open-source macOS pointer companion), Jev as a decision layer for Grok Bot, HA-Jev for Home Assistant, Jev as a Judge (LangChain-shared evaluator with repository), Vercel's free-until-25-September announcement, the founder's 17-minute talk, two content-analysis demos (1,315 posts for $0.086; 100,000 posts in 20.4 s), a real-time trading bot, elvis on Jev as a harness primitive, a where-it-is-free platform list, two X Articles (Japanese 20-use-case roundup, Nate Herk's 12 use cases), 20 tips, a browser skill demo and the AppitStudio directory.
GitHub metadata now covers all 106 referenced repositories.
The new badge and the new-entry count are gone; the summary shows the total only.

Records touched: Jev as a Judge: decision model vs LLM judges for evals, TipTour: a local pointer companion for macOS, Jev as a decision layer for Grok Bot, HA-Jev: Home Assistant integration, Vercel AI Gateway: Jev free until 25 September, Viral post analyser: 100,000 X posts in 20.4 seconds.

MVP v1.2 2026-09-20

Repositories page with GitHub metadata

/repos/: every GitHub repository the index points at (106) in one sortable table — stars, language, license and last push from the public GitHub API — with search, category and language filters stored in the query string. Machine-readable copy at /data/repos.json.
npm run refresh-github pulls the metadata into data/github.json (cache; 60 requests/hour unauthenticated, GITHUB_TOKEN raises it). Detail pages show stars, language and last push for entries with a repository.
Repos added to the navigation; the open-repositories count on the home page links to the new page.

Batch 6 2026-09-20

13 high-reach X posts and articles (50k+ views)

13 records added (230 total) from an X search restricted to posts above 50,000 views: seven X Articles (setup roadmaps by codila, Daniel Ch and Movez; Matt Van Horn's nine-projects roundup; Japanese and Chinese explainers; a Chinese deployment list; an HFT trading guide), the HiringCafe resume-relevance benchmark on a 2.5M-MAU app, DeRonin's where-the-100x-is post, DCVC's 72-hours post, and two launch-week threads.
Original X posts linked on existing GitHub-first entries: kev (Jared Palmer) and the TypeSafe launch blog; typesafe-mario gained the source walkthrough post.

Records touched: HiringCafe: resume × job relevance benchmark, Jev Engineering: a 10-step setup roadmap, WTF is Jev? Nine things people are already building, How to use Jev, and where the 100x actually is, DCVC: 72 hours after launch, Got Jev — now what? A real deployment list.

Batch 5 2026-09-20

49 entries from a second research pass, posters, author filters

49 records added (217 total). Official: the jev-1.13 known-weaknesses page and the status page. Framework and observability integrations: Pydantic AI, LiteLLM compaction, Braintrust, Arize OpenInference, Vercel's explainer and AI SDK provider. Six more open models and replicas, nine benchmark repositories and tests, thirteen tools and demos, eight builder posts, three analysis pieces.
Foreman and TypeAR now link their repositories (thruwire/foreman, TypeAR-AI/TypeAR). openjev is now SemIf; the entry links the live demo and the Hacker News thread.
52 entries show a reduced-size poster from the original post or article, with a media note on the Methodology page.
Directory: 48 cards at a time with a Show more button; author names on cards filter the directory in place; every detail page lists other entries by the same author; sitemap lastmod follows each entry's verification date.
Snapshot date moved to 2026-09-20; the new badge marks entries added in the latest snapshot.

Records touched: Jev 1.13 known weaknesses ("jaggedness"), Pydantic AI: first-class TypeSafeModel, LiteLLM: Jev-driven context compaction and pass-through, open-jev-deberta-v3-large (Kotoba Labs), Jev as a reranker: graded-relevance eval, Jev Paper Judge, Foreman: a supervisor for coding agents, TypeAR: Jev-inspired type-safe decoding.

Batch 4 2026-09-19

61 more entries: the rest of the verified research candidates

61 records added (168 total): repositories from the 18 September link catalog (jev-ultrafast, jev-review, jev-router, mobile-jev, jevlike, official skills repo, awesome-jev-by-typesafe); six further replicas (system-one-gemma, daseinlabs/open-jev, jev-on-a-laptop, reflex, DiffusionGemma open-jev, TypeAR); Rust, Ruby and TypeScript clients; database and routing integrations (pg_typesafe, duckdb-jev, tiershift, prismhq/jev-router); guardrails (jev-shield, Jevegis, agentgateway example); directories; live demos with stated authorship; 19 builder posts from X; press and analysis including Forbes, DataCamp, AlphaSignal and two Japanese pieces.

Records touched: Jev Ultrafast (Browser Use), Official agent skills repository, Jevegis: guardrails API on Jev, jev-shield: semantic MCP firewall, reflex: a small open decision model with a WebGPU build, Forbes: the $200M startup fixing AI overconfidence.

Batch 3 2026-09-19

58 new entries from a web, GitHub and X research pass

58 records added (107 total): official SDKs, the evals dashboard, the funding press release and pre-launch essay; Jev on Vercel AI Gateway, OpenRouter, Cloudflare Workers AI and Netlify; five open replicas (kev, openjev-sglang, simple-jev, system-one-open, sgoedecke/system-one); agent tooling (toolgate, jev-mcp, jev-rules, hono-jev-router, n8n node, LlamaIndex reranker, neo4jev, blink, Unclutter); four independent benchmark repositories and four vendor/builder tests; press and analysis; twelve builder demos from X.
New collection: Benchmarks & independent tests. Existing collections extended.
Author correction: the driving-simulation post is by @jpschroeder.

Records touched: TypeSafe workflow evals dashboard, kev: a tiny Jev-like you can train on a MacBook, openjev-sglang, Unclutter: Jev-powered clutter blocker, Jev as an LLM router: a mostly negative result, Near Here: Jev vs Mistral vs Gemini on event validation, Vercel's fx safety classifier on Jev, Driving-simulation demo with Jev, cobanov/awesome-jev.

MVP v1.1 2026-09-19

Collections, feeds, New badge, link health

Eight curated collections under /collections/ (Start here, Browser & computer use, Agent harnesses, Search & ranking, Coding, Games & simulations, Local & replica models, Directories & roundups); every member page links back to its collections.
Atom feed (/feed.xml) and JSON Feed 1.1 (/feed.json) with evidence label, kind and destination fields.
Snapshot-relative new badge on cards and a new-entry count in the summary.
npm run check-links: link-health report.

MVP v1 2026-09-19

Permanent pages, explicit destinations, validated data

Canonical data moved from data/sources.js to validated data/sources.json with a documented schema.
Every record separates the source (sourceUrl / xUrl) from product destinations (projectUrl, demoUrl, repoUrl, docsUrl). The primary button opens the project; the X post has its own button.
Evidence labels: official / source-backed / author-demo / commentary, documented on the Methodology page.
Each of the 49 records has a permanent page under /projects/{slug}/ with related entries and a Last verified date.
Search, type, evidence, tag filters and sort live in the query string and survive refresh.
Deterministic Node build with validation and tests; generated sitemap.xml, robots.txt and 404 page.
Links added from the research link catalog: fast-jev-compaction repository, Jev Search live demo, TypeSafe console and evals, LangChain integration docs, the SemIf redirect for openjev, and original X posts for GitHub-first entries.

Records touched: Instant context compaction, Jev Search, Jev-like model on a home GPU, Introducing System One Models and Jev, Harness design with Jev, Early Jev GitHub projects, The early Jev projects thread.

Prototype 2026-09-19

Research index prototype

Single-page index of 49 entries from the two research batches (2026-09-18 and 2026-09-19) with type filters, full-text search and a detail dialog.
Warm paper, black rules, orange emphasis and acid-green evidence markers established as the visual system.